Efficient touch based localization through submodularity

Abstract

Many robotic systems deal with uncertainty by performing a sequence of information gathering actions. In this work, we focus on the problem of efficiently constructing such a sequence by drawing an explicit connection to submodularity. Ideally, we would like a method that finds the optimal sequence, taking the minimum amount of time while providing sufficient information. Finding this sequence, however, is generally intractable. As a result, many well-established methods select actions greedily. Surprisingly, this often performs well. Our work first explains this high performance -we note a commonly used metric, reduction of Shannon entropy, is submodular under certain assumptions, rendering the greedy solution comparable to the optimal plan in the offline setting. However, reacting online to observations can increase performance. Recently developed notions of adaptive submodularity provide guarantees for a greedy algorithm in this online setting. In this work, we develop new methods based on adaptive submodularity for selecting a sequence of information gathering actions online. In addition to providing guarantees, we can capitalize on submodularity to attain additional computational speedups. We demonstrate the effectiveness of these methods in simulation and on a robot.

Pose estimation for contact manipulation with manifold particle filters

“…Recent work uses probabilistic methods for the tactile localization of immovable objects [25][26][27]. These systems produce a number of distinct touch actions that provide information about the object pose.…”

Pose estimation for planar contact manipulation with manifold particle filters

“…The problem of particle starvation when using contact sensors in a particle filter have been recognized several times in the literature (Gadeyne et al 2005, Zhang & Trinkle 2012, Zhang 2013). This problem is commonly addressed by "smoothing" the observation model with artificial noise that spreads contact observations over a non-infinitesimal, full-dimensional region of the state space (Zhang 2013, Javdani et al 2013, Zhang & Trinkle 2012, Corcoran & Platt 2010. This approach-while sometimes effective-scales poorly to high-resolution sensors and discards the most important property of contact sensors: the difference between contact and no-contact.…”

“…This approach is commonly implemented by executing a sequence of move-until-touch actions (Petrovskaya & Khatib 2011, Javdani et al 2013, Hebert et al 2013, Hsiao 2009) that localize the object within some tolerance, then execute an open-loop trajectory to achieve a grasp. These techniques generally assume that the object does not move (Javdani et al 2013, Petrovskaya & Khatib 2011 or use a simple motion model that causes actions to "bump" the object by a small amount (Hsiao 2009). The MPF solves a fundamentally different problem: it estimates the pose of an object during manipulation and does not plan any actions.…”

Adaptive Informative Path Planning in Metric Spaces

“…It touches on several important research topics, which contain one or two, but not all three elements. If we focus on information gathering only and ignore robot movement cost, IPP becomes sensor placement, view planning, or ODT, which admits efficient solutions through, e.g., submodular optimization, in both nonadaptive [15] and adaptive settings [7,13]. If we account for movement cost, there are several nonadaptive algorithms with performance guarantee (e.g., [12,19]).…”

“…• A mobile manipulator moves around and senses an object with laser range finders [18] or tactile sensors [13] in order to estimate the object pose for grasping.…”